Update PII detector model card

Upload README.md with huggingface_hub

Add entity-type list (40 types x 11 domains) to model card

Upload README.md with huggingface_hub

Ship Group-B cue tier (amount .16->.77; internal F1 .691->.727); refresh card

Refresh v8 numbers w/ context tier (internal .691) + development-notes log

Ship context-cued ID decode tier (internal F1 .641->.691; Group-A IDs ~0->.6-.75)

Note v9 outcome (evaluated, not published)

Add external-model comparison (GLiNER, Piiranha, OpenMed, OpenAI-PF, LFM demo) + plot

Add all-candidate-models comparison table + plot

Decode upgrade: boundary-snap + precision guards (exact-F1 0.361->0.429, proxy ->0.829)

v8: best on internal+ShieldFlow+most public; 40-type + hybrid decode

v7: 40-type multilingual PII + hybrid decode

Add internal-eval (v1) results + plots

corrected proxy-gate numbers (crosswalk fix) + reproduced q4 baseline

v4: +v7 proxy in-distribution (gate F1 .60->.72), 32 types, #1 on all 4 clean held

leaderboard: fair GLiNER prompting

v3a: +public train splits (in-distribution) -> #1 on all 4 clean benchmarks

v2: +EU langs/legal/format-diversity, judged-1M+2M-v2 data

add leaderboard plot + embed in card

pii-detect interim (ckpt-4000, epoch 0.23) + public-eval leaderboard